Unverified50% confidenceFactExact time
The gradient of standard MTP contains two components: a KL self-distillation signal and a noise residual.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSoft-MTP decouples the KL self-distillation signal from the noise residual, so that during gradient backpropagation the gradient flows only through the KL self-distillation part.80% similarUnverifiedThe core setup of the study is that the signal and the noise each have independent sparsity parameters while the signal-to-noise ratio remains of constant order—the signal strength does not grow as the matrix size increases.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1118515API
curl https://kongchang.com/api/v1/knowledge/claims/1118515MCP
get_claim(id=1118515)